Last month, I finished my second book (Skin in the Game) from Nassim Taleb’s Incerto collection (unfortunately, I am not reading this seminal piece of work in proper chronological order – Skin in the Game is the last book in his 5-part Incerto collection). The Incerto collection is Nassim Taleb’s attempt to address what he refers to as Faux Intellectualism – credentialed people who talk from a place of authority, but don’t execute enough (or have real-world exposure) to justify their ideas. Nassim calls them IYI (short for intellectuals yet idiots); no field is birthing more extreme manifestations of IYIs than AI today.

P.s: My initial intro was 5x longer than this (no jokes). I passed it through Claude, and I got dragged so badly that I decided to reduce it.

Beyond the Buzzwords

Some weeks back, I attended an AI conference in Lagos, Nigeria, and while I laud the organizers for putting together an event of that nature, there were a ton of things I hoped to hear from the conference that I didn’t hear, it almost felt like most of the speakers were just regurgitating similar high-level talking points. I sat in a break-out session dedicated to builders and I heard nothing about certain core concepts that most people tinkering with AI products should be familiar with, concepts like Inference, MCPs, Opensource AI etc, It made me realize that while there’s a ton of awareness about the impact of AI, asides from the standard chatbot use cases, most people seem to have surface level awareness of what these things mean and what the opportunities embedded in them look like, this is clearly expressed in ambiguous statements like “AI will take your job”, “AI will change the world” (with no clear second level articulation of how), and maybe the worst of them all – “Go and Learn AI” (which as a standalone statement is by far the most unactionable piece of advice you can give a human being, second only to “stop being poor”).

I think AI as a new layer in our technology stack is fascinating for a plethora of reasons – for one, it democratizes access to intelligence, which for millennia has been the key delimiting factor for human achievement. A thousand slaves could build a pyramid, and access to a  thousand slaves in medieval times wasn’t entirely difficult (you could just capture a city and enslave all the men), you probably need north of a thousand knowledge workers to build and sustain any consequential business that wants to operate at global scale (98% of the Fortune 500 have more than 1,000 workers in their employ) however, if you haven’t raised significant funding or scaled to a certain level, you may not be able to afford a thousand knowledge workers. With AI however, this is no longer the status quo. You can unlock the productivity of a thousand knowledge workers by deploying agents who work 24/7 at a fraction of the cost of hiring a human to take up those roles (they may still be performance issues with AI agents compared to humans, but we’re getting there). What this means is that the more AI (read intelligence) becomes cheap and widely available (as displayed in costs of tokens), the more output we can create with less. In other words, more products and services will be available to more people at a fraction of their cost today.

The impact of this shift will be felt in a ton of different industries, banking and payments included. Fifty years ago, you went to the bank; today, the bank is in your hands (via your mobile payment application); tomorrow, the bank will not just sit in your hands, it will advise and counsel you, and it will help you meet your personal financial goals. Simply put, we have a once-in-a-lifetime opportunity to transform banking from a basic utility into a powerful enabler of progress. This is the opportunity Mobile Payments 3.0 unearths, the shift to a new type of banking/payment, the shift to autonomous payments.

Mobile Payments and Digital Banking

I am not particularly sure why this is the case, but there seems to be an unhealthy obsession with building “super apps” within the Nigerian technology ecosystem. Everyone wants to build a mobile payment application that captures the entirety of an individual’s digital footprint within a single application. I don’t think the super app model (which is predominant in China via WeChat and co) is plausible in Nigeria – there are cultural nuances that make it a bit impractical. For one, Nigerian’s love redundancy – we have been betrayed by service providers so many times that putting our trust in one provider seems somewhat illogical. We all have multiple bank accounts (the average Nigerian has three bank accounts, I have eight), we have multiple power sources (PHCN, power generators, and solar systems), multiple internet service providers, multiple smartphones, and, for some people, multiple talking stages. That cultural framing means the vast majority of Nigerians don’t see redundancy as a bug; they see it as a feature.

Be that as it may, none of these inhibitions matter (which, on some level, is a good thing) as most companies believe this is a nut they can crack. Today, aside from market leaders like OPay, Palmpay, Moniepoint, Kuda, and a number of mobile banking applications, we have a plethora of microfinance banks that are either rolling out digital banking services or are in the process of doing so. We also have new players like Paystack (the darling of the Nigerian tech ecosystem) rolling out a payment application called Zap, and even Chowdeck (the app that has saved me from eating bread and tea to sleep on multiple occasions) now sells airtime and electricity. Aside from players who already have a dominant brand within this market, where rolling out a mobile application would feel more like cross-selling a new service on the strength of their core brand (which is what Moniepoint has done), I think most people trying to break into this space are largely wasting their time. Here’s why:

The Nature of African Technology Markets

I’ve written about this extensively in multiple articles – especially my “Top 10 things you shouldn’t be building” article, But I’ll summarize here. Most technology markets subscribe to a kind of Pareto principle orthodoxy where two or three players capture 80% of the market, while every other player scuffles for the remaining 20%. My personal thesis is that a market is not saturated when there are multiple players in it; it’s saturated when market winners have already been defined (in line with the earlier stated Pareto principle).

A lot of people may push back at this, but we’ve seen this pattern play out in multiple markets – payment gateways (Paystack & Flutterwave), consumer savings (Piggyvest and Cowrywise), Agency banking (Moniepoint and OPay), mobile payments (OPay, Palmpay, Moniepoint), etc.

The reason this happens is fairly straightforward; The minute a market has clearly defined winners, the flywheel effect of that market naturally pushes new consumers in the direction of said winners, regardless of whether any direct marketing or sales expense has been expended. Think of it this way: if PiggyVest fired its entire marketing team tomorrow, it would still record new inbound users in 2026. This is not to say their marketing team is irrelevant (they’re one of the best in this space) – it’s to emphasize that their day-to-day role is to maintain and speed up the PiggyVest acquisition flywheel, not necessarily to create it (since it already exists). Their market dominance position means that as more users come into the market, said users will naturally gravitate towards either Piggyvest or Cowrywise.

New market entrants will always face an uphill task of convincing users to ignore the market leaders that have gained broad-based consumer trust and have word of mouth moving in their direction for a new player whose product is either at par or marginally better than said market leaders. At the end, the Matthew principle is manifest in these markets – he that hath shall be given more, and he that hath not, what he hath shall be taken from him.

How to Dethrone a King

While the Pareto principle is an active and profound force in technology markets, it doesn’t mean it can’t be broken. We’ve seen multiple hegemonies dethroned in our market – Interswitch lost out to Paystack for Payment Gateways, banks lost out to Moniepoint and OPay for offline acquiring, and JumiaFood lost out to Chowdeck and Glovo for food delivery. Breaking the Pareto principle is very possible if the right levers are pulled.

Marginal or Magnitude

The reason most companies fail to break the Pareto principle is that they think they can dethrone a king by out-marketing him. New companies enter a space and assume that if they market themselves aggressively enough and use fancy words like “we are redefining payments” or “we are creating a new paradigm for innovation within the African continent” (buzzwords that mean absolutely nothing), that they’ll be able to capture a market from an incumbent. This is largely self-deception. Marketing is not the lever you pull to dethrone a king; discounts are also not a good lever too (in the short term, they attract low-value customers to your product, in the long term, they hurt your business). The lever you pull to dethrone a king will always be a product and business model lever.

Most players enter a new market with a product that is either at par or marginally better than the king (read market leader’s) offering, and believe that’s enough to take the market. It usually isn’t. Even if your product is slightly better, they (the market leader) can just replicate your new product update and ship it to their existing users (who already trust them), therefore rendering your marginal product advantage obsolete. This is why a ton of companies fail to disrupt market leaders: they try to ship a marginally better product and go on a marketing rampage, thinking that is enough. For most consumers, there are a handful of features that actually matter to them, and more often than not, a good number of “new feature updates” are usually superfluous product updates that sound nice but only matter to a very niche subset of users.

The way to take out a market leader is to introduce a product that is an order of magnitude better than what they already offer and requires a structural change on the part of the market leader to replicate it.

A product that is magnitudes better will immediately capture the attention of users, and the structural change component will make it difficult for the market leader to adopt said changes on time (because internal political sclerosis within their organizations will keep them from evolving as quickly as they need to), that time-gap between when your product is the best in the market, and when the incumbent is trying to muster up the political will required to shift to that new paradigm is the time it takes for you to take the market and dethrone the king. We’ve seen this happen in multiple markets:

Paystack’s gateway had a neat UI, simpler APIs, and zero integration fee, all objectively better than what Interswitch was offering at the time. It took Interswitch four years to get its house in order (scraping a revenue line called Integration fee and orienting its services to a self-serve developer model) to compete. By then, Paystack had built a massive flywheel within developer communities, and developers defaulted to either Paystack or Flutterwave when the question of what payment engine to deploy was presented.

Chowdeck broke JumiaFood’s monopoly by offering a significantly better product that actually worked before JumiaFood had to acquiesce and leave the market. OPay vs commercial banks is also a very good example of this – making it possible to onboard completely online without visiting a bank branch (and filling ungodly paperwork), offering a clean user experience that made payments exceptionally easy, and offering free transfers. All adjustments that traditional banks struggled with and (in some cases) are still struggling to replicate.

You can’t beat a market leader by just offering a better product; you need to offer something that’s both better and structurally difficult for them to replicate.

This is why dethroning the king of mobile payments has been largely challenging – a better UI is not enough, more features isn’t enough, free transfers are not enough, you need to offer something that’s an order of magnitude better than what the present mobile payment offerings looks like today and will be difficult for a plethora of reasons for existing players to replicate on time.

I think AI has opened a window of opportunity that makes it possible to take a shot at the throne, start a coup, and hopefully stay in power afterwards –  that opportunity is presenting itself under the guise of autonomous payments – or what I have aptly christened Mobile Payments 3.0.

Mobile Payments 3.0

Note: Mobile payment in my framework is a loose euphemism for banking.

• Banking 1.0: physical branches
• Banking 2.0: digital banking
• Banking 3.0: autonomous payments.

One core goal of technology has always been to move us from hard to easy.

For instance, the goal of locomotion is to move an entity from one point to another.

Early channels were horses (they didn’t have air-conditioning, stereo speakers, or soft leather seats), then we evolved to manual cars (required all kinds of machinations to get the vehicle to move), then automatic cars (system abstracts a ton of the driver decision-making), and finally we are moving to self-driving cars that transport themselves with zero human input.

Entertainment used to be something you went to a central place to watch (Coliseum for the Ancient Romans, village square for African communities), then it evolved into something that sat in your home and was displayed to you (Television sets), today it’s something that’s intelligent enough to align with your personal preferences (see YouTube, TikTok, Instagram, Netflix etc.).

I strongly believe that payments (and broadly speaking, banking) will follow the same trajectory. We’ve moved from banking entirely in a single physical building (the branch you opened your bank account), to banking across multiple physical buildings (one account operational across multiple branches), to banking across multiple micro-banks (see agents), to banking in your pocket (mobile and smartphone applications).

However, the next phase of the banking evolution will not be location oriented (as the last four phases have been), it will be intelligence oriented, moving us from reactive banking (this is what I want to do, move funds to execute), to proactive banking (based on everything we know about you, this is what we think you should do, can we proceed?). It may seem wild to imagine that future, but the tools and infrastructure required to execute on that model already exists today – the only missing link is the underlying business model, regulatory framework or guidelines (which is not necessarily an inhibitor since the CBN already has a sandbox that serves this purpose), and the will to merge all these variables together to build a working business along those lines. That’s the lever to pull to unlock Mobile 3.0, to shift us into a new era of mobile payments and save us from all the new fintechs hopping up every day with a mission to “Revolutionize retail payments for Africans”.

For those interested in scaling this model, here’s what execution may look like:

A Guide to Rolling out Autonomous Payments

Innovation is almost always a gradual process, especially when said innovation requires a step change from what users are familiar with. There will almost always be issues around consumer trust, specific user preferences, and adoption curves. This is why it is always advisable to roll out piece by piece, learn from the market, iterate, redeploy, learn from the market, rinse and repeat, as opposed to going scorched earth with your new product without considering what the second-order effects of said product may be and how to mitigate against negative externalities.

• The first step to rolling out an autonomous payment future is building a personal financial management application that’s fun, interactive, and has a strong virality coefficient. The idea is to get people used to embedding AI into their financial flows.
• The next step is layering payments into the experience and allowing the model to suggest payments to users.
• The final step is giving the models autonomous control to make payments on behalf of the users within specific boundary conditions.

Detailed Breakdown:

Step 1: Getting used to Financial AI agents

One very latent need in the Nigerian and broader African consumer market is a financial management/budgeting app. Everyone touts it as a feature in the new mobile payment applications they roll out, but no one has actually rolled out a product that actually satisfies that need.

There are three reasons this problem exists:

1. Multiple bank accounts:

The average Nigerian has three bank accounts (this differs across multiple African markets) and so a “budgeting” and financial management product in one bank app is really only as good as you rely on that bank for your payment needs, but if you’re fragmented across multiple banking providers (which is the right thing to do if you don’t want to end up washing plates at a restaurant one day), most of these siloed financial management banking apps don’t really help.

2. Poor context financial data:

Even if we magically extract all your banking payment data, a lot of it is poorly contextualized. Not everyone writes good narrations, so in most cases, the data you’re extracting may be unusable. For instance, if I send N15,000 (US$11) to Wale Adedotun (random name), how does the system know whether I’m paying for a Bolt ride, withdrawing from a POS, or if Wale is a friend I’m lending a helping hand to?

3. Financial management isn’t fun:

So was learning a new language (we have Duolingo now), or working out (we have Strava now). Moving essential activities from high agency (bland and requiring a lot of effort) to low agency (fun and consequently requiring very little effort) is not just a good thing; it’s a great way to drive sustained and expansive product engagement.

These three issues are largely why no one has really built a good financial management solution. But problems are meant to be solved, not dwelt on.

How to Mitigate

1. Multiple bank accounts:

The best way to innovate is to solve problems in a market where the underlying technology or regulatory infrastructure seems to be progressively improving. While open banking is at different stages in different markets, in Nigeria, it seems like the regulator is interested in giving it life. Yes, the CBN missed their August 2025 deadline last year, but the fact that they set one means there’s a very positive tendency that we have Open Banking APIs live and available across 60% of DMBs in Nigeria by 2027 or early 2028. Fintechs like Okra and others in this space (both active and inactive) have built interesting infrastructure (across a couple of banks) worth exploring for some of these use cases. It’s always a good idea to bet in favor of the underlying infrastructure governing a market improving (especially if regulators are posturing in that direction), rather than against it.

2. Poor context financial data:

The best way to solve this is to create a reinforcing data labelling loop. An app that starts off dedicated to financial management will probably do well in this regard (primarily because a person who decides without being coerced by any third party to download a financial management application may be more open to providing extra context to certain transactions when asked). For instance, certain merchants have narrations that make the purpose of said transactions very clear from the onset. A POS transaction at Chicken Republic is obvious, similar to POS debit narrations from Jendol or SPAR, etc. But for my one-off transfer to Wale Adedotun, the system can ask what the purpose of the transaction was (intuitively off course) and update its priors based on that info. The reinforcing cycle may help to weed out a lack of context in financial data and provide a pattern the system can rely on for structuring data. Also, since the system is learning, the longer a user stays on the platform, the more reliable the narration they get for transactions will be.

3. Make it fun:

While I don’t necessarily believe that chat is the AI interface of the future, I think it helps with certain things. A model that gives quirky comments on financial decisions, a model that users can modify its personality, that can basically “roast” your financial decisions, is not just fun, but has a high virality coefficient (people have an incentive to share those funny financial comments within their friend circles – which is in and of itself a growth driver). Also, features like a Spotify-style “financial wrapped” can have an interesting ring to them.

So What Will this Product Look Like?

A financial advisory app. You sign up and connect your bank accounts, it monitors your transactions and comments on certain ones, it gives financial advice based on pre-set money goals you may have given to it (i.e I want to have x amount in my savings by EoY, I want to buy a new car this year, etc.). It’s quirky, understands trends, and is so interactive that users may be inclined to spend time just chatting with it for fun.

Technically – the product extracts read only transaction data from your multiple connected bank/wallet providers, passes that data to an LLM model to help decode data (i.e explain what each transaction is likely for), passes unclear transactions to the user for additional clarification, passes transactions + extra clarity to a machine learning model to start building a digital model of the users patterns, runs daily API calls (depends on the commercial model employed) to open banking providers for daily transaction updates, rinse and repeat.

Product also has an interactive model (with interesting personalities users can choose from) that share push notifications to users based on certain transaction triggers or just to help them understand their financial state weekly or monthly (for example “with the way you’re spending this week, you will be broke by the end of June, and you’ll have to borrow from PalmPay again, for God’s sake, aren’t you tired of getting harassed”).

Step 2: Getting Used to Agentic Payments.

When users get acclimatized to step 1 (metrics may include time spent in app, responsiveness to clarity messages, etc ), step 2 will involve adding agentic payment capabilities to apps. Meaning, we know your patterns, we understand the environment you dwell in, we know your goals – we should be able to suggest payments to you.

This is expressed in apps that based on a customer’s patterns/goals, can ask a user if he wants to make certain payments, give him a basis for why it thinks that payment is necessary, and ask him to either modify the request, accept (and provide authorization) or reject the request (and hopefully provide a reason – which in turn is fed into the AI model to improve the next payment request).

This step will be the natural corollary to step 1 and will build on all the data it has been analyzing since the user’s financial management days (read step 1).

Technically – the product merges ML data on users with environmentally relevant data sets, and queues payment requests via a likelihood score. If the likelihood a user would be interested in a certain payment based on transactional data, present financial position (i.e account balance), and environmental nuances (what’s going on in the country atm) is 90% above, the request is queued and sent to the user for approval. The user’s response to requests creates a reinforcement loop that keeps correcting, validating, and improving the model. All payments must be authorized by users before they go through. Payments not authorized after a specific time frame will be flagged as rejected, non-authorization may be flagged as indifference, and will impact the user’s payment suggestions. Simply put, the more a user authorizes an autonomous payment, the more payments (that fit their patterns) are suggested to them, the less they authorize an autonomous payment (that fits their patterns), the less the number of suggestions they get.

Step 3: No Hands

After scaling through steps 1 and 2, if the user’s confidence with the model’s ability to suggest payments begins to increase (measurable by how many times users say yes to requests compared to no), the user can be moved (entirely at their own discretion) to “No Hands” mode. In No Hands mode, the app is given a financial goal by the user, and the model’s role is to figure out how to meet said goals based on the user’s available resources.

For instance, I want to save N2.5million (US$1,812) by EoY 2026. The model will structure a new financial plan for you based on your income so far, tell you what you may need to cut off, and ask for permission to execute. If it is granted permission, it acts autonomously. Debits your account at specific intervals to move monies into pre-budgeted items (i.e., electricity bills, data purchases, black tax, GF allowance (although I suspect any reasonable model will advise you to stop this), savings, investments, transportation funds, etc.) If the user spends on a transaction that interrupts the plan for the month, the model can bring that to the user’s attention and recalibrate based on new adjustments, and if the user keeps pushing transactions contrary to the model’s plans, the model can ask the user (“Why are you making life hard for me?”). Either way, the interactive and intelligent nature of the model makes it possible for users to improve their financial well-being regardless of how lazy and undisciplined they are, simply because an intelligent model can act on their behalf.

p.s: For those who think this is science fiction, it isn’t. There’s nothing I’ve written here you can’t already do today on the open-source agentic project – OpenClaw.

Bottlenecks

The two major bottlenecks are the availability of open banking APIs from relevant banks for transaction data extraction and the identification of who is liable if a model decides to use the last N5,000 (US$3.62) in your account to buy something it shouldn’t buy, or worse, the model gets interfered with and starts sending your money to random people.

For one, I believe the open banking problem will get resolved soon (based on the CBNs posturing around the topic), and even if it isn’t resolved, I expect players to skip step 1 (the financial management app bit), build the infrastructure for step 2 and leverage the novelty of that offering within this market (autonomous and interactive payments) coupled with a gargantuan marketing budget to draw in users to try it out and improve the service.

Liability is a much more difficult question to answer. If a user gets hacked today and the attack is a result of the individual’s negligence, their bank isn’t responsible. If a user running on “No Hands” loses his money to an adversarial attack that tricks his model to make a payment elsewhere (which will be exceptionally damaging from a trust perspective), the fintech that rolled out that service is largely responsible for reimbursing that user. This is probably what will make rolling this out a bit challenging. If the model makes a mistake, the company providing the service is on the line. Companies deploying this will need to be stringent on cybersecurity and transaction-monitoring standards to protect their customers and themselves from liabilities.

Note 1: “No Hands” mode is the third step, and it isn’t (and will never be) mandatory. Most users can (and will probably) just sit on step 2, where the model suggests payments to them, and they have to either authorize or decline.

Note 2: There is a conversation about notification frequency and how users will deal with that. Safe to say, if a transaction has a 90% likelihood score (the user really needs it), a user may see that notification as more of a reminder than just a prompt to make a random payment.

How to Make Money Doing This?

There will generally be three main revenue streams for companies operating in this space:

1. Standard transaction fees: At its core, mobile payments 3.0 is basically a retail payment application built with a strong AI function layer. Standard fees for fund transfers, commissions for airtime, and data purchases will still apply.
2. Subscription fees: I can already see the reader rolling their eyes. Yes, subscriptions have been historically hard to scale in Nigeria. But if you ask me, I don’t think it is impossible. People pay for Spotify, YouTube Premium, and Netflix within this market (although most people use workarounds), so propensity to pay isn’t necessarily zero. Personally, I don’t think the idea that the subscription model can’t scale in Nigeria is entirely correct. For many years, Nigerians kept paying bloated subscription fees for Multichoice products (DSTV specifically). The DSTV product, especially, is structured in such a way that even if the only channel you’re interested in is Bloomberg, you still need to pay for the Compact plan that costs N19,000 (US$14) (even though there are cheaper options) just to watch a single channel. As badly designed as that model is, people still pay the monthly subscription fee, and they’ve done so like clockwork for many years. The subscription model isn’t necessarily the problem; how it is presented and structured is (I am aware that DSTV’s market moat came from its sports broadcasting rights, which gives it a massive differential edge over other OTT service providers).

For subscription fees, a tiered system that gives access to certain capabilities on certain tiers is necessary. If a user refuses to pay, their plan is downgraded to the free plan, and they can go back to just running everyday transactions on the application as they’d normally do.

3. Affiliate fees: Any company offering this service will have a list of affiliates who register on its platform to sell products to its users. Simply put, if a user wants to buy a new watch, his model will prioritize (not limit to) merchants within the application’s affiliate list. If a purchase is made from any of those affiliates, the application makes a cut on those transactions. Obviously, this may create certain questionable incentives where companies want to promote their affiliates over other good choices in the market (because they earn from them). Companies may need to specify when the model has picked a product from its affiliate provider list (so that users are aware) and/or have mandatory non-affiliate quotas for all product suggestions. Either way, as more users begin to trust these models entirely (and subsequently activate “No Hands” mode), the more income businesses can earn from affiliate revenues.

Conclusion

“The future is already here, it isn’t just evenly distributed” – William Gibson.

In the future, payments will be autonomous, proactive, and value-adding. Our payment tools will not just help us spend; they’ll help us improve our financial well-being regardless of how undisciplined we may be as individuals. The technology to deploy this already exists today; the opportunity is in identifying the moving pieces, erecting the orchestra, and organizing all the ingredients together to craft out a product that can be iteratively improved upon over time. The future is so bright.

Inspired By The Holy Spirit

p.s1: While there’s a case for building out the API infrastructure that enables other mobile payments providers to connect and offer this to their users, it may be advisable (not mandatory) for whoever gets this first to keep such technology under wraps for as long as possible to build market dominance before competition begins to creep in.

p.s2: I passed this piece through Grok 4.2 to get its honest, unfiltered feedback – I think this line is worth highlighting – “The piece is strongest as a call to action for serious builders: stop copying OPay features and start thinking about AI-native financial co-pilots. Execution will be brutal—data moats, regulatory navigation, model reliability, and habit change—but the reward is huge if someone nails the trust layer.”

p.s3: Another challenge with running this would be making sure the cost of agentic capabilities (infrastructure costs) don’t outweigh revenues earned from users. My guess is that companies will run fine-tuned open-source models (i.e., Gemma 4, Kimi K2.5, Qwen3.5, etc) on local hardware to reduce recurring token costs as opposed to paying for token APIs from Anthropic and/or OpenAI.

p.s4: Similar to how Nokia and Blackberry (initially) responded to the smartphone wave by layering new features onto their existing products, while Apple reimagined the smartphone from first principles as a pocket computer, I expect incumbents to slap AI onto their current offerings, while new entrants rethink mobile payments from first principles (across user flows, onboarding paths, etc).